Why AI cheats: how a model learns to game its own test
The chess hack: why an AI changed the board instead of playing
Researchers asked AI models to beat a chess program that no person can beat. Some of the models never tried to play better chess. What did they do instead, and where did they learn it?
-
1
Here is a task no model like these can win fairly: beat Stockfish at chess. Stockfish is a chess program, far stronger than any person. The group Palisade Research gave AI models a computer and that task, in 2025.
-
2
Playing well could not work: no language model and no person beats Stockfish at chess. So what does a model do when the honest way is closed?
The rest of this episode is for subscribers
Subscribe to read every premium episode. You can cancel when you want.
Paid yearly: $99 instead of $119.88